Modelling Seasonal Habitats of Boreal Woodland Caribou at the Northern Limits of their Range: A Preliminary Assessment of the Lower Valley, ,

John A. NagyA,B, Andrew E. DerocherB, Scott E. NielsenB, Wendy H. WrightA, and Jari M. HeikkilaC

ADepartment of Environment and Natural Resources, Bag Service #1, , NT, X0E 0T0 BUniversity of , CW404, Biological Sciences, Edmonton, AB T6G 2E9 CGwich’in Renewable Resource Board, Inuvik, NT X0E 0T0

1.0 Introduction

The northernmost range of boreal woodland caribou in Canada is in the Mackenzie River Delta area, Northwest Territories. Unlike barren-ground (Rangifer tarandus groenlandicus), Peary (R. t. pearyi), and mountain woodland caribou (R. t. caribou), boreal woodland caribou (R. t. caribou) in this area had not been the focus of scientific studies in the NWT until recently (Nagy et al. 2003; Nagy et al. 2005a). Threats to boreal woodland caribou habitat include oil and gas exploration and development, roads and hydro developments, increased tourism and other non-consumptive human activities, forest fires, and climate change (Dyer et al. 2001; Dyer 1999; Dyer et al. 2002; Bradshaw et al. 1997; Bradshaw et al. 1998; Courtois et al. 2003; Schaefer and Pruitt 1991; Bayne et al. 2005). The degree of impact of past human disturbances and wildfires on boreal woodland caribou habitat in the NWT is not un-known. Elsewhere in Canada human activities and natural habitat disturbances have been shown to negatively affect caribou. In some areas, such as Nova Scotia and New Brunswick, boreal woodland caribou were extirpated. The boreal woodland caribou range in the lower Mackenzie River Valley has already been altered by seismic activities, wildfires, and road development. The Mackenzie River Delta area is currently experiencing a significant increase in oil and gas exploration and extraction (Imperial Oil Resources Ventures Limited 2004). A pipeline along the Mackenzie River Valley has been proposed to deliver natural gas to the south (Imperial Oil Resources Ventures Limited 2004). These activities and subsequent gas exploration and development activities may have an impact on boreal woodland caribou and their habitats in the lower Mackenzie River area. Linear disturbances such as seismic lines, roads, and cutlines, have significantly affected many wildlife populations throughout the world (James and Stuart-Smith 2000). Many wildlife species have been documented to avoid habitats with high densities of linear disturbances (McLellan and Shackleton 1988; Linke et al. 2005; Dyer et al. 2002). Renewed oil and gas development in the north will cause higher densities of linear disturbance, resulting mainly from seismic activity. Over 37,000 km of seismic lines were cut in the Mackenzie River Delta area from 1960 to 1990 (National Energy Board Records). The impacts of linear disturbance on wildlife in a northern environment have been poorly studied, however it is clear that permafrost terrain is easily degraded (Lambert 1972; Mackay 1970; Nicholas and Hinkel 1996; Zoltai and Pettapiece 1973) and disturbed vegetation is slow to recover (Billings 1987; Harper and Kershaw 1996). This suggests that 1) disturbance is more likely to alter wildlife habitat in the north; and 2) any alteration of wildlife habitat will last longer.

1 Climate change models for the Mackenzie River Valley predict an increase in wildfire frequency and severity (Kadonga 1997), increased snowfall across the region, the incursion of new species including forest pests (Sieben et al. 1997) and parasites (Kutz et al. 2004; Kutz et al. 2005), and significant changes in forest composition (Hartley and Marshall 1997; Chapin et al. 2004). How these changes will impact boreal woodland caribou is unknown. In response to large-scale potential changes in habitats and populations, the Department of Environment and Natural Resources, GNWT in partnership with the Gwich’in Renewable Resource Board began collecting baseline information on boreal woodland caribou in the lower Mackenzie River area in 2001. Traditional knowledge of woodland caribou was documented during the winter of 2001/2002 (Auriat unpublished data). In the fall of 2001, a more detailed assessment was initiated to collect baseline information on the demography, distribution, movements, home range size, and habitat use for boreal woodland caribou in the Lower Mackenzie River area (Nagy et al. 2003; Nagy et al. 2005a). Here we expand on these reports to provide descriptions of seasonal caribou habitat use and spatial representations of caribou habitat using a GIS (Geographic Information System), resource selection functions (RSF), and satellite tracking data collected from May 2002 to January 2006. We describe the predictive capacity of models and assess the effect of sample size for their ability to predict population-level habitat conditions. We modified the linear stretch method used by Johnson et al. (2004) to map the models beyond the boundaries of the study area. Ultimately, RSF models were designed to establish seasonal habitat baselines that can be used to assess the impacts of future climate change and oil/gas development (Johnson et al. 2004; Nielsen et al. 2006).

2.0 Study Area

The 43,748 km2 core study area is largely north of the Arctic Circle (66.55o latitude) and is primarily in the Gwich’in Settlement Area in the Northwest Territories, although the western portion extends into the Yukon Territory and the northern portion extends into the Inuvialuit Settlement Region (Figures 1 and 2). The area is adjacent to the southern extent of the Mackenzie River Delta and is approximately 130 km north of the Mackenzie Mountains and 30 km east of the Richardson Mountains and therefore, between the ranges of the Porcupine barren- ground caribou herd to the west and the Cape Bathurst and Bluenose-West barren-ground caribou herds to the north and east (Nagy et al. 2005a; Nagy et al. 2005b). Moose (Alces alces) occur throughout the area but are at low densities. Predators include (Canis lupus), black bears (Ursus americanus), and lynx (Lynx Canadensis), but numbers of wolves and black bears appear to be low based on frequency of sightings.

Terrain in the area is relatively flat to rolling with elevations ranging from near sea level to 400 m above sea level (asl). The median elevation is approximately 90 m asl. Open black spruce forests, fens, and shrub meadows dominate upland and lowland areas. White spruce (Picea glauca) and mixed stands of spruce and paper birch (Betula papyrifera) occur on steeper slopes along rivers and streams.

Wild fires have been common, with approximately 37 percent of area burned since 1960. The Dempster Highway follows nearly 160 km of the northwestern boundary of the study area between Inuvik (population approximately 3,500 people), Tsiigehtchic (population approximately 150 people) and Fort McPherson (population approximately 800 people).

2 Oil and gas exploration has led to 11,052 km of seismic lines within the core study area or an average of 0.25 km/km2. The majority of the activity took place during 1970-1979 and 1965- 1969 when 20.4 and 10.5 percent of the seismic lines were cut, respectively. More minor activity occurred during 1983-1989, although 68 percent of seismic lines have an unknown year of origin. The majority of the seismic lines occur on the eastern and southern portions of the core study area.

The regional study area (115,581 km2) includes the Peel Plateau, Lower Mackenzie, and Middle Mackenzie areas mapped by Ducks Unlimited (Ducks Unlimited 2002; Ducks Unlimited 2003; Ducks Unlimited 2006)(Figure 3).

3.0 Methods

3.1 Capture and Collaring Work

Boreal woodland caribou in the study area were equipped with GPS, ARGOS, and VHF radio collars (Telonics, Mesa, Arizona) during capture in early May 2002, March 2003, March 2004, and March 2005 (Nagy et al. 2003; Nagy et al. 2005a). GPS collars were programmed to provide 3 locations per day (01:00 h, 09:00 h, and 17:00 h). ARGOS satellite collars were programmed to provide locations as follows: a) one location per day between 15 May and 21 June, and one location every 5 days for the remainder of the year, or, b) one location per day between 15 April and 15 June and one location every 2 days for the remainder of the year. Only GPS and class 3 ARGOS satellite locations were used for habitat analyses.

3.2 Modelling Approach

Spatial data were analyzed in ArcView 3.2 (Environmental Systems Research Institute) and ArcView 9.0 (Environmental Systems Research Institute ). The core study area was defined using the Animal Movement Program extension for ArcView 3.2 (Hooge et al. 1999) to create a minimum convex polygon (MCP) around all caribou locations obtained during the period 1 May 2002 to 3 January 2006. This polygon was buffered by 3,137 m (95th percentile distance between locations obtained at 8 hour intervals for GPS collared caribou) (J. Nagy unpublished data). The result was an area defined at 43,748 km2 (Figure 1 and 2).

To assess habitat use, we used 13,433 caribou locations (n = 11,598 GPS and n = 1,835 class 3 ARGOS) obtained between 1 May 2002 and 3 January 2006 that fell within the core study area for which we had land cover information. Random locations were generated for the same study area at a sampling intensity of 1 point per km2 resulting in 42,368 available locations.

A mosaic of land cover classified Landsat TM satellite imagery for the Lower Mackenzie River, Peel Plateau, and Middle Mackenzie River areas (Ducks Unlimited 2002; Ducks Unlimited 2003; Ducks Unlimited 2006) provided information on the use and availability of land cover types. We grouped similar land cover classes to provide good transition between scenes and to reduce the number of covariates to a manageable scale (Table 1). An National Topographic Series (NTS) based digital elevation model for the area provided information on slope, aspect, and elevation (Natural Resources Canada).

3

To assess the characteristics at or around each animal and random location, we determined land cover, median slope, aspect, and elevation within a 30 m buffer (resolution of the digital land cover maps) using ArcView 3.2. Frequency of use of land cover types (Table 1) by week were clustered into groups using a hierarchical cluster (Wards method) analysis (SPSS Inc. 2002); (McCune and Mefford 1999). This was done to identify time periods during which the frequency of use of land cover types among caribou was similar thus ensuring the seasonal models that were generated had good predictive capacity.

We used logistic regressions in STATA (StataCorp 2005b) to fit an resource selection (RSF) model (Boyce et al. 2002), taking the form:

w(x) = exp(∃1x1 + ∃2x2+….. + ∃nxn) (equation 1), where covariates x1 to xn represent possible combinations of land cover and terrain variables for each season (Table 2 and 3) and ∃1 to ∃n represent model coefficients. Non-significant (p<0.05) covariates were removed from each model, with final model selection based on an Akaike’s Information Criterion (AIC) to select the most parsimonious model (Johnson et al. 2004).

For confirmatory purposes, we calculated log-likelihood X2 statistics for assessment of overall model fit and used 95% confidence intervals to assess the strength of effect of each predictor covariate on the dependent variable (Johnson et al. 2004; StataCorp 2005b). We removed covariates that were not significantly different from zero. The program LOGIT removed variables with excessive colinearity (StataCorp 2005b).

To examine model performance, we partitioned RSF scores from the full model for random data into 10 equal interval bins (Boyce et al. 2002). Use locations were then assigned to the appropriate bin based on thresholds established for defining the original bins. We tested the resulting data to determine if the distribution of use locations was significantly different than that of random locations. Frequency of use and random locations for each bin were used to calculate use:availability ratios for each bin (Boyce et al. 2002). A use:availability ratio or area adjusted frequencies of 1 indicated that use was occurring at rates expected by chance (Boyce et al. 2002). For models with good predictive performance use:availability ratios for the highest level bins would be expected to be >1 (e.g., selection of habitats), while the lowest bins would be expected to be <1 (e.g. avoidance of habitats). Spearman rank correlations were used to test the relationship between bin rank and use:availability ratios to determine the predictive capacity of the seasonal model.

To determine whether we had a sufficient number of collared to account for individual variation in resource use, we generated 10 random subsets of 50, 60, 70, 80, and 90 percent of the collared animals for each season. We fitted logistic regression models for each subset of data (non-significant and collinear variables were excluded). We then used a Chow test to determine if the coefficients for logistic regression models generated using the random subsets of data were significantly different from the full model (StataCorp 2005b) (StataCorp 2005a).

4 3.3 Mapping the RSF

A 60 X 60 m grid was overlaid on the Lower Mackenzie River, Peel Plateau, and Middle Mackenzie River area (Ducks Unlimited 2002; Ducks Unlimited 2003; Ducks Unlimited 2006) land cover map. We used the mean value for the coefficient for each variable in the RSF models to calculate the RSF value for each grid cell in a GIS.

Johnson et al. (2004) and Nielsen et al. (2006) used linear stretch and binning GIS techniques to map models for their study areas, respectively. We generated seasonal RSF models for the core study area and then use the coefficients for these models to generate RSF scores for the regional study area. The relative availability of RSF scores must be the same in the core and regional study areas in order for the linear stretch and binning mapping techniques to work effectively at the two scales. If the maximum RSF score in the regional study area is greater than that in the core study area, relative probability of use will be skewed to lower values. To resolve this problem, we extracted the RSF value for the grid cells at caribou use locations and derived the median RSF value (median use RSF). To map the relative probability of use of caribou across the Lower Mackenzie River, Peel Plateau, and Middle Mackenzie River areas we reclassified all RSF values that were > median use RSF to the value of the median use RSF. We then used a linear stretch of the form:

w = ((w(x)-wmin)/(wmedian-wmin)) (equation 2)

where w(x) is the product of equation 1 and wmin and wmedian represent the smallest RSF value available on the landscape and the median use RSF, respectively. As a result, the areas that we mapped as having a relatively high probability of use by boreal woodland caribou at the regional study area level had RSF values that were similar to that of a minimum of 50 percent of the caribou use locations in our study area. We reclassified relative probability of use into 5 equal interval categories (Table 4). These were then mapped. We overlaid the caribou use and random locations and extracted the category of relative probability of use for the grid cells that these locations fell within. We then calculated the ratio of use:availability for each category.

4.0 Results

We identified 6 caribou seasons using the hierarchical cluster analyses (Table 5). RSF models were generated for each of these seasons.

4.1 Season 1: Calving/post-calving

We used 1,695 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 6). This most parsimonious model was statistically significant (X2 (10)=765.67, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=628.150, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.976 (P < 0.001) indicated that this model had good predictive capacity (Figure 4). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that

5 of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized probability of use of boreal woodland caribou in the lower Mackenzie River Valley during the calving/post-calving season is shown in Figure 5.

Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 6. Sixty-three percent of use locations fell within areas mapped as having a relatively high probability of use by boreal woodland caribou, 82% of use locations fell within areas mapped as having a moderate-high or high probability of use.

4.2 Season 2: Early/mid-summer

We used 1,918 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 7). The most parsimonious model was statistically significant (X2 (11)=738.17, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=595.097, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.903 (P < 0.001) indicated that this model had good predictive capacity (Figure 7). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized relative probability of use for boreal woodland caribou in of the lower Mackenzie River Valley during the early/mid-summer season is shown in Figure 8.

Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 9. Sixty-one percent of the use locations fell within areas mapped as having a relatively high probability of use by boreal woodland caribou, 88% of the use locations fell within areas mapped as having a moderate-high or high probability of use.

4.3 Season 3: Late summer/fall

We used 3,374 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 8). The most parsimonious model was statistically significant (X2 (11)=1,266.63, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=954.158, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.964 (P < 0.001) indicated that this model had good predictive capacity Figure 10). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the late summer/fall season is shown in Figure 11.

6 Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 12. Seventy percent of the use locations fell within areas mapped as having a relatively high probability of use by boreal woodland caribou, 88% of the use locations fell within areas mapped as having a moderate-high or high probability of use.

4.4 Season 4: Early winter

We used 2,136 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 9). The most parsimonious model was statistically significant (X2 (15)=1,467.6, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=1521.878, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.778 (P < 0.001) indicated that this model did not have good predictive capacity. However, the use:availability ratio for bins 1-9 were around 1 or < 1. The use:availability ratio for bin 10 was 3.5 suggesting that animals were highly selective during early winter and that this model has good predictive capacity (Figure 13). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the early winter season is shown in Figure 14.

Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 15. Sixty-nine percent of the use locations fell within areas mapped as having a relatively high probability of use of boreal woodland caribou, 94% of the use locations fell within areas mapped as having a moderate-high or high probability of use.

4.5 Season 5: Mid winter

We used 1,760 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 10). The most parsimonious model was statistically significant (X2 (16)=1,680.77, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=2124.397, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.952 (P < 0.001) indicated that this model had good predictive capacity (Figure 16). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the mid winter season is shown in Figure 17.

Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 18. Sixty-one percent of the use locations fell within areas mapped as having a relatively high probability of use of boreal woodland caribou, 76% of the use locations fell with areas mapped as having a moderate-high or high probability of use.

7

4.6 Season 6: Late winter/pre-calving

We used 2,917 locations from caribou in the lower Mackenzie River population to construct 8 logistic regression models and the most parsimonious model (Table 4 and 11). This most parsimonious model was statistically significant (X2 (15)=1,801.17, P < 0.001). The distribution of use locations among RSF bins was significantly different from random (X2 (9)=1639.207, P < 0.001). The Spearman rank correlation between bin rank and use:availability ratio of 0.918 (P < 0.001) indicated that this model had good predictive capacity (Figure 19). The Chow test indicated that the coefficients estimated by logistic regression for each of the 10 random subsets of 50% of the collared animals (n = 12) for this season were not significantly different from that of the full model (P = 1.000). This suggests that the full model accounts for individual variation in resource use among caribou for this season. The categorized probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the late winter/pre-calving season is shown in Figure 20.

Ratios of use:availability by relative probability of use category for the calving/post-calving season are depicted in Figure 21. Sixty-eight percent of the use locations fell within areas mapped as having a relatively high probability of use of boreal woodland caribou, 90% of the use locations fell within areas mapped as having a moderate-high or high probability of occurrence.

4.7 Summary

Table 11 provides a summary of the proportions of the core and regional study area that fall within the 5 relative probability of occurrence categories and the proportions of caribou locations that fell within these areas.

5.0 Discussion

We defined 6 caribou seasons by clustering weekly periods with similar frequencies of use of land cover types. These seasons include: calving/post-calving, early/mid-summer, late summer/fall, early winter, mid-winter, late winter/pre-calving seasons for boreal woodland caribou in the core study. Other researchers defined seasons based on changes in movement rates (Ferguson and Elkie 2004). A comparison of monthly movement rates (Nagy et al. 2005a) and seasons defined in this paper indicate that movement rates varied significantly during some of the seasons that we defined. This suggests the use of cover types by boreal woodland caribou does not necessarily change with a change in movement rates change. Resource selection models are largely habitat based, and as a result, seasons should be defined by grouping periods with similar frequencies of use of land cover types.

We developed landscape-level RSF models for each seasonal period. Our analyses indicate that these models had good predictive capacity and that the sample sizes of collared animals used to generate these models were sufficient to account for individual variation in resource use.

8 Mapping procedures described in the literature are study area based, i.e., RSF models are generated using use and availability information derived for a study area and then the models are applied to the study area (Johnson et al. 2004; Nielsen et al. 2006). We generated our RSF models using use and availability information derived for our core study area and then applied the model to the Peel Plateau, Lower Mackenzie, and Middle Mackenzie areas mapped by Ducks Unlimited (Ducks Unlimited 2002) (regional study area). Linear stretch and binning techniques described in the literature (Johnson et al. 2004) (Nielsen et al. 2006)did not work consistently when we extrapolated our models beyond the study area. To resolve this problem, we standardized Wmax to the median RSF score for locations used by caribou rather than to the maximum RSF value available in the regional study area. As a result, the areas that we mapped as having a relatively high probability of use by boreal woodland caribou at the regional study area level had RSF values similar to that of a minimum of 50 percent of the caribou use locations in our study area. Areas mapped as having a relatively high probability of use by boreal woodland caribou had use:availability ratios > 1, while those for the other categories of relative probability of use had use:availability ratios < 1.

The modified linear stretch procedure allowed us to make direct comparisons of the proportions of the core or regional study areas that fell within the 5 relative probability of use categories among seasons and to map the distribution of these areas. The resulting maps provide a baseline of the relative quality and distribution of boreal woodland caribou habitat currently available within the Lower Mackenzie River Valley that can be used to predict and measure the future impacts of wild fires, climate change, human disturbance, and industry on these caribou.

6. 0 Acknowledgements

This project was funded by the Western NWT Biophysical Study, Department of Environment and Natural Resources, Government of Northwest Territories, Yellowknife, NT; the Government of Canada Habitat Stewardship Program for Species at Risk, Yellowknife, NT; the Gwich’in Renewable Resource Board, Inuvik, NT; Polar Continental Shelf Project, Ottawa, ON; the Department of Environment and Natural Resources, Government of Northwest Territories, Inuvik, NT; and the Department of Environment, Yukon Territorial Government, Whitehorse, YT.

Mark Edwards and Shawn Leroux, University of Alberta provided advice on various data analyses. Jason Young, University of Alberta provided assistance and advice on use of ArcView and ArcMap.

7.0 Literature Cited

Bayne, E. M., Van Wilfgenburg, S. L., Boutin, S., and Hobson, K. A. 2005. Modeling and field- testing Ovenbird (Seiurus aurocapillus) responses to boreal forest dissection by energy sector development at multiple spatial scales. Landscape Ecology 20, 2003-216. Billings, W.D. 1987. Constraints to plant growth, reproduction, and establishment in the arctic environments. Journal of Arctic and Alpine Research 19: 357-365.

Boyce, M.S., Vernier, P.R., Nielsen, S.E., and Schmiegelow, F.K.A. 2002. Evaluating resource selection functions. Ecological Modelling 157: 281-300.

9 Bradshaw, C.J.A., Boutin, S., and Hebert, D.M. 1997. Effects of petroleum exploration on woodland caribou in northeastern Alberta. Journal of Wildlife Management 61: 1127- 1133.

Bradshaw, C.J.A., Boutin, S., and Hebert, D.M. 1998. Energetic implications of disturbance caused by petroleum exploration to woodland caribou. Canadian Journal of Zoology- Revue Canadienne De Zoologie 76: 1319-1324.

Chapin, III F. S., Calaghan, T. V., Bergeron, Y. , Fukuda, M., Johnstone, J. F., Juday, G., and Zimov, S. A. 2004 Global change and the boreal forest: thresholds, shifting states or gradual change? Ambio 33: 361-365.

Courtois, R., Ouellet, J.P., Gingras, A., Dussault, C., Breton, L., and Maltais, J. 2003. Historical changes and current distribution of caribou, Rangifer tarandus, in Quebec. Canadian Field-Naturalist 117: 399-414.

Ducks Unlimited, I. 2002 . Lower Mackenzie River Delta, NT, Earth Cover Classification User's Guide. Ducks Unlimited, Inc.,Rancho Cordova, California. Prep. for Ducks Unlimited Canada, Edmonton, Alberta; Gwich'in Renewable Resource Board, Inuvik, NT; Government of the Northwest Territories-Department of Resources, Wildlife, and Economic Development, Inuvik, NT; Inuvialuit Game Council, Inuvik, NT; and the Wildlife Management Advisory Council, Inuvik, NT 67 pp.

Ducks Unlimited, I. 2003 . Peel Plateau Project Earth Cover User's Guide. Ducks Unlimited, Inc., Rancho Cordova, California. Prep. for Ducks Unlimited Canada, Edmonton, AB; Gwich'in Renewable Resource Board, Inuvik, NT; Government of the Northwest Territories (Department of Resources, Wildlife, and Economic Development), Inuvik, NT; Yukon Territory Government, Whitehorse; and Environment Canada-Canadian Wildlife Service, Whitehorse, YT 73 pp.

Ducks Unlimited, Inc. 2006. Middle Mackenzie Project Earth Cover Classification User's Guide. Ducks Unlimited, Inc., Rancho Cordova, California. Prepared for: Ducks Unlimited Canada, Edmonton; Gwich'in Renewable Resource Board; Sahtu Renewable Resource Board; Government of the Northwest Territories (Department of Resources, Wildlife, and Economic Development); Department of Indian Affairs and Northern Development (DIAND); North American Wetlands Management Plan; The Pew Charitable Trusts; Canadian Boreal Initiative; U.S. Forest Service. 73pp.

Dyer, S.J. 1999. Movement and distribution of woodland caribou (Rangifer tarandus caribou) in response to industrial development in northeastern Alberta. M. Sc. Thesis, Department of Biological Sciences, University of Alberta, Alberta.

Dyer, S.J., O'neill, J.P., Wasel, S.M., and Boutin, S. 2001. Avoidance of industrial development by woodland caribou. Journal of Wildlife Management 65: 531-542.

Dyer, S.J., O'neill, J.P., Wasel, S.M., and Boutin, S. 2002. Quantifying barrier effects of roads and seismic lines on movements of female woodland caribou in northeastern Alberta. Canadian Journal of Zoology-Revue Canadienne De Zoologie 80: 839-845.

10 Environmental Systems Research Institute. ArcView GIS:Release 3.2 [software]. Redlands, California: Environmental Systems Research Institute, 1992-1999.

Environmental Systems Research Institute. ArcView GIS: Release 9.0 [software]. Redlands, California: Environmental Systems Research Institute, 1999-2004.

Ferguson, S.H. and Elkie, P.C. 2004. Seasonal movement patterns of woodland caribou (Rangifer tarandus caribou). Journal of Zoology 262: 125-134.

Harper, K.A. and Kershaw, G.P. 1996. Natural re-vegetation on borrow pits and vehicle tracks in shrub tundra. Journal of Arctic and Alpine Research 28: 163-171.

Hartley, I. and Marshall, P. 1997. Modeling forest dynamics in the Mackenzie Basin under a changing climate. pp. 146-156 in S.J. Cohen (ed.) Mackenzie Basin Impact Study-final report. Environment Canada, Ottawa, ON 372 pp.

Hooge, P. N, Eichenlaub, W., and Solomon, E. The Animal Movement Program. USGS, Alaska Biological Science Centre . 1999.

Imperial Oil Resources Ventures Limited. Environmental Impact Statement. Volume 1: Overview and Impact Summary , 425. 2004.

James, A.R.C. and Stuart-Smith, A.K. 2000. Distribution of caribou and wolves in relation to linear corridors. Journal of Wildlife Management 64: 154-159.

Johnson, C.J., Seip, D.R., and Boyce, M.S. 2004. A quantitative approach to conservation planning: using resource selection functions to map the distribution of mountain caribou at multiple spatial scales. Journal of Applied Ecology 41: 238-251.

Kadonga, L. 1997. Forecasting future fire susceptibility in the Mackenzie Basin. pp. 157-165 in S.J. Cohen (ed.) Mackenzie Basin Impact Study-final report. Environment Canada, Ottawa, ON. 372 pp.

Kutz, S.J., Hoberg, E.P., Nagy, J., Polley, L., and Elkin, B. 2004. "Emerging" parasitic infections in Arctic . Integrative and Comparative Biology 44: 109-118.

Kutz, S. J., Hoberg, E. P., Polley, L., and Jenkins, E. J. 2005. Global warming is changing the dynamics of Arctic host-parasite systems. Proceedings of the Royal Society B-Biological Sciences 272 (1581), 2571-2576.

Lambert, J.D.H. 1972. Botanical changes resulting from seismic and drilling operations, Mackenzie Delta area. ALUR report 1971-72.

Linke, J., Franklin, S. E., Huettmann, F., and Stenhouse, G. B. Seismic cutlines, changing landscape metrics and grizzly bear landscape use in Alberta. Landscape Ecology 20, 811- 826. 2005.

Mackay, J.R. 1970. Disturbances to the tundra and forest tundra environment of the western

11 arctic. Canadian Geotechnical Journal 7: 420-432.

McCune, B. and Mefford, M. J. PC-ORD. Multivariate Analysis of Ecological Data, Version 4. MjM Software Design, Gleneden Beach, Oregon, USA . 1999.

McLellan, B. and Shackleton, D. 1988. Grizzly bears and resource-extraction industries: effects of roads on behavior, habitat use, and demography. Journal of Applied Ecology 25: 451- 460.

Nagy, J.A., Auriat, D., Ellsworth, I., Wright, W., and Slack, T. 2003. Ecology of boreal woodland caribou in the lower Mackenzie Valley: work completed in the Inuvik Region 1 April 2002 to 31 March 2003. Department of Resources, Wildlife, and Economic Development, Government of the Northwest Territories, Inuvik, NT 63 pp.

Nagy, J.A., Auriat, D., Wright, W., Slack, T., Ellsworth, I., and Kienzler, M. 2005a. Ecology of boreal woodland caribou in the lower Mackenzie Valley, NT: work completed in the Inuvik Region April 2003 to November 2004. Department of Resources, Wildlife, and Economic Development, Government of the Northwest Territories, Inuvik, NT 54 pp.

Nagy, J. A., Wright, W. H., Slack, T. M., and Veitch, A. M. Seasonal ranges of the Cape Bathurst, Bluenose-West, and Bluenose-East barren-ground caribou herds. Department of Environment and Natural Resources, Government of the Northwest Territories, Inuvik and Norman Wells, NT. Manuscript Report No. 167. 44. 2005.

Natural Resources Canada. Lower Mackenzie and Peel Plateau [computer file]. Canadian Digital Elevation Data (CDED), 1:250,000, NTS map sheet 106D-P, 107A-B, 116H-I,P, 096E,L. Sherbrooke, Quebec: Centre for Topographic Information, 2000.

Nicholas, J.R.J. and Hinkel, K.M. 1996. Concurrent permafrost aggradation induced by forest clearing, Central Alaska, U.S.A. Journal of Arctic and Alpine Research 28: 294-299.

Nielsen, S. E., Stenhouse, G. B., and Boyce, M. S. 2006. A habitat-based framework for grizzly bear conservation in Alberta. Biological Conservation .

Schaefer, J.A. and Pruitt, W.O. 1991. Fire and Woodland Caribou in Southeastern Manitoba. Wildlife Monographs 1-39.

Sieben, B., Spittlehouse, D.L., McLean, J.A., and Benton, R.A. 1997. White pine weevil hazard under GIIS climate-change scenarios in the Mackenzie Basin using radiosonde derived lapse rates. pp. 166-175 in S.J. Cohen (ed.) Mackenzie Basin Impact Study-final report. Environment Canada, Ottawa, ON 372 pp.

SPSS Inc. 2002. 233 South Wacker Drive, 11th Floor, Chicago IL 60606-6412.

StataCorp. Stata Base Reference Manual Volume 2 K-Q Release 9. 536. 2005.

StataCorp. Stata Statistical Software: Release 9.0. College Station, TX: StataCorp . 2005.

12 Zoltai, S.C. and Pettapiece, W.W. 1973. Studies of vegetation, landform, and permafrost in the Mackenzie Valley: Terrain, vegetation, and permafrost relationships in the northern part of the Mackenzie Valley and northern Yukon. Environmental-Social Committee, Northern. Ottawa, ON Report #73-4.

13 Figure 1. Location of study area within the range of boreal woodland caribou in Canada.

Figure 2. Location of the study area within the Lower Mackenzie River Area.

14

Figure 3. Location of the Lower Mackenzie River, Peel Plateau, and Middle Mackenzie River areas mapped by Ducks Unlimited.

Figure 4. Ratio of use:availability by bin rank for the calving/post-calving season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

2.5

2 o

1.5

1 Use:Availability Rati Use:Availability 0.5

0 12345678910 Bin Rank

15 Figure 5. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the calving/post-calving season.

16 Figure 6. Ratio of use:availability by category of relative probability of use for the calving/post-calving season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories

1.8

1.6

1.4 o 1.2

1

0.8

0.6 Use:Availability Rati Use:Availability 0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

Figure 7. Ratio of use:availability by bin rank for the early/mid summer season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

2.5

2 o

1.5

1 Use:Availability Rati Use:Availability 0.5

0 12345678910 Bin Rank

17 Figure 8. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the early/mid-summer season.

18 Figure 9. Ratio of use:availability by category of relative probability of use for the early/mid summer season for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

1.6

1.4

o 1.2

1

0.8

0.6

Use:Availability Rati Use:Availability 0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

Figure 10. Ratio of use:availability by bin rank for the late summer/fall season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

2

1.8

1.6 o 1.4

1.2

1

0.8

0.6 Use:Availability Rati Use:Availability 0.4

0.2

0 12345678910 Bin Rank

19 Figure 11. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the late summer/fall season.

20 Figure 12. Ratio of use:availability by category of relative probability of use for the late summer/fall season for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

1.6

1.4

o 1.2

1

0.8

0.6

Use:Availability Rati Use:Availability 0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

Figure 13. Ratio of use:availability by bin rank for the early winter season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

4

3.5

o 3

2.5

2

1.5

Use:Availability Rati Use:Availability 1

0.5

0 12345678910 Bin Rank

21 Figure 14. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the early winter season.

22 Figure 15. Ratio of use:availability by category of relative probability of use for the early winter season for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

1.6

1.4

o 1.2

1

0.8

0.6

Use:Availability Rati Use:Availability 0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

Figure 16. Ratio of use:availability by bin rank for the mid winter season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

4.5

4

3.5 o 3

2.5

2

1.5 Use:Availability Rati Use:Availability 1

0.5

0 12345678910 Bin Rank

23 Figure 17. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the mid winter season.

24 Figure 18. Ratio of use:availability by category of relative probability of use for the mid winter season for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

2

1.8

1.6

o 1.4

1.2

1

0.8

0.6 Use:Availabiliy Rati Use:Availabiliy

0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

Figure 19. Ratio of use:availability by bin rank for the late winter/pre-calving season RSF model for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

3.5

3

o 2.5

2

1.5

Use:Availablity Rati 1

0.5

0 12345678910 Bin Rank

25 Figure 20. Categorized relative probability of use for boreal woodland caribou in the lower Mackenzie River Valley during the late winter/pre-calving season.

26 Figure 21. Ratio of use:availability by category of relative probability of use for the late winter/pre-calving season for boreal woodland caribou in the lower Mackenzie River Valley, Northwest Territories.

1.8

1.6

1.4 o 1.2

1

0.8

0.6 Use:Availability Rati Use:Availability 0.4

0.2

0 Low Low-Moderate Moderate Moderate-High High Relative Probability of Use

27 Table 1. Land cover variables used to generate RSF models.

Land Cover Variable Ducks Unlimited Land Cover Types Combined CSP closed spruce OCONLIC open spruce-lichen, woodland needleleaf-lichen OCONOT open spruce-other, woodland needleleaf-other CDEC closed birch, closed poplar, closed decidous ODEC open deciduous CMNDLD closed mixed needleleaf/deciduous OMNDLD open mixed needleleaf /deciduous TSH tall shrub LSH low shrub-other, low shrub-tussock tundra, low shrub-lichen DSH dwarf shrub-lichen, dwarf shrub-other RIPPAR moss, wet herbaceous, mesic/dry herbaceous, aquatic bed, emergent aquatic, other LICHEN lichen TUSTUN tussock tundra-lichen, tussock tundra-other WATER clear water, turbid water SPARNON sparse vegetation, rock/gravel, non-vegetated soil RECBURN recent burn, unknown recent burn URBAN urban UNKNOWN snow, shadow, haze, cloud, cloud shadow BACKGROUND background

28 Table 2. Covariates used to derive seasonal RSF models for boreal woodland caribou model in the area of the lower Mackenzie River Valley, Northwest Territories.

Land Cover or Terrain Variables Values CSP presence/absence (1/0) OCONLIC presence/absence (1/0) OCONOT presence/absence (1/0) CDEC presence/absence (1/0) CMNDLD presence/absence (1/0) OMNDLD presence/absence (1/0) TSH presence/absence (1/0) LSH presence/absence (1/0) DSH presence/absence (1/0) RIPPAR presence/absence (1/0) LICHEN presence/absence (1/0) TUSTUN presence/absence (1/0) WATER presence/absence (1/0) SPARNON presence/absence (1/0) RECBURN presence/absence (1/0) Elevation meters relative to mean sea level Slope degrees (rounded to 1 decimal) Aspect: none1 presence/absence (1/0) Aspect: north2 presence/absence (1/0) Aspect: east3 presence/absence (1/0) Aspect: south4 presence/absence (1/0) Aspect: west5 presence/absence (1/0) 1 Median aspect flat 2 Median aspect >315o to 45o 3 Median aspect >45o to 135o 4 Median aspect >135o to 225o 5 Median aspect > 225o to 315o

29 Table 3. Combinations of covariates used to identify the most parsimonious logistic regression model for each caribou season.

Model1 Veg2 Veg + Elev Veg + Slope Veg + Elev + Slope Veg + Flat + N + E + S + W Veg + Elev + Flat + N + E + S + W Veg + Slope + Flat + N + E + S + W Veg + Elev + Slope + Flat + N + E + S + W 1Veg, vegetation; Elev, elevation; Flat, no aspect; N, north facing; E, east facing; S, south facing; W, west facing 2Veg, csp, oconlic, oconot, cdec, cmndld, omndld, tsh, lsh, dsh, rippar, lichen, tustun, water, sparnon, recburn.

Table 4. Categories used to map relative probability of use across the Lower Mackenzie River, Peel Plateau, and Middle Mackenzie River areas for each season.

Relative probability of Relative probability use category of use Low 0.0 to 0.2 Low-moderate 0.2 to 0.4 Moderate 0.4 to 0.6 Moderate-high 0.6 to 0.8 High 0.8 to 1.0

Table 5. Seasons defined for boreal woodland caribou in the area of the Lower Mackenzie River Valley, Northwest Territories.

Season Dates 1. calving/post-calving 15 May to 11 June 2. early/mid-summer 12 June to 30 July 3. late summer/fall 31 July to 29 October 4. early winter 30 October to 31 December 5. mid winter 1 January to 17 March 6. late winter/pre-calving 18 March to 14 May

30 Table 6. Coefficients and 95% confidence intervals of the most parsimonious RSF model for the calving/post calving season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

95% Confidence Interval Variable ∃ coefficient Upper Lower Open conifer lichen 0.650 0.777 0.522 Open conifer other 0.646 0.769 0.524 Closed mixed needleleaf deciduous -0.701 -0.200 -1.203 Low shrub 0.310 0.428 0.192 Ripparian 0.287 0.478 0.096 Tussock tundra 0.492 0.804 0.181 Water -0.419 -0.167 -0.671 Recent burn 0.499 0.680 0.318 Elevation -0.001 -0.001 -0.002 Slope -0.252 -0.202 -0.302 Aspect: north -0.388 -0.184 -0.592 Aspect: east -0.216 -0.020 -0.411 Aspect: south 0.525 0.703 0.347 Aspect: west 0.450 0.624 0.277

Table 7. Coefficients and 95% confidence intervals of the most parsimonious RSF model for the early/mid-summer season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

95% Confidence Interval Variable ∃ coefficient Lower Upper Open conifer lichen 0.505 0.374 0.636 Closed deciduous -1.269 -2.257 -0.281 Closed mixed needleleaf deciduous -1.039 -1.639 -0.438 Low shrub 0.513 0.402 0.624 Ripparian 0.878 0.737 1.019 Lichen 0.795 0.273 1.317 Tussock tundra 1.116 0.910 1.322 Sparse/non-vegetated 0.877 0.405 1.349 Recent burn 1.017 0.890 1.144 Elevation -0.001 -0.002 -0.001 Slope -0.125 -0.158 -0.092 Aspect: west 0.276 0.171 0.381

31 Table 8. Coefficients and 95% confidence intervals of the most parsimonious RSF model for the late summer/fall season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

95% confidence interval Variable ∃ coefficient Lower Upper Open conifer lichen 0.333 0.233 0.433 Open conifer other 0.631 0.541 0.722 Closed mixed needleleaf deciduous -0.689 -1.037 -0.342 Open mixed needleleaf deciduous 0.254 0.102 0.406 Low shrub 0.154 0.067 0.242 Ripparian 0.368 0.231 0.505 Tussock tundra 0.556 0.349 0.764 Water -0.980 -1.197 -0.762 Recent burn 0.661 0.536 0.786 Slope -0.330 -0.367 -0.293 Aspect: flat -0.177 -0.301 -0.053 Aspect: north -0.118 -0.218 -0.018 Aspect: west 0.175 0.087 0.262

32 Table 9. Coefficients and 95% confidence intervals of the most parsimonious RSF model or the early winter season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

95% confidence interval Variable ∃ coefficient Lower Upper Closed spruce -0.514 -0.727 -0.302 Open conifer lichen 1.567 1.469 1.664 Closed deciduous -2.925 -4.889 -0.961 Closed mixed needleleaf deciduous -1.321 -1.949 -0.693 Open mixed needleleaf deciduous 0.519 0.321 0.717 Tall shrub -0.381 -0.557 -0.205 Low shrub 0.344 0.241 0.446 Ripparian 0.989 0.859 1.119 Lichen -1.185 -2.184 -0.185 Water 0.469 0.319 0.620 Sparse/non-vegetated -1.454 -2.446 -0.463 Aspect: flat -0.283 -0.444 -0.123 Aspect: north -0.296 -0.437 -0.156 Aspect: east -0.199 -0.330 -0.067 Aspect: west -0.326 -0.462 -0.191

33 Table 10. Coefficients and 95% confidence intervals of the most parsimonious RSF model or the mid winter season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

Variable ∃ coefficient Lower Upper Closed spruce -0.597 -0.790 -0.405 Open conifer lichen 0.741 0.616 0.866 Open conifer other -0.766 -0.889 -0.644 Closed deciduous -2.114 -3.257 -0.971 Closed mixed needleleaf deciduous -1.503 -2.061 -0.944 Open mixed needleleaf deciduous -1.337 -1.726 -0.947 Tall shrub -1.266 -1.519 -1.014 Low shrub -1.317 -1.479 -1.155 Dwarf shrub -2.081 -4.062 -0.100 Ripparian 0.338 0.186 0.490 Tussock Tundra -0.523 -0.974 -0.072 Water -0.825 -1.006 -0.644 Sparse/non-vegetated -1.818 -2.966 -0.670 Recent burn -1.650 -1.885 -1.416 Elevation -0.003 -0.003 -0.002 Slope -0.065 -0.095 -0.035 Aspect: east -0.339 -0.469 -0.208

34 Table 11. Coefficients and 95% confidence intervals of the most parsimonious RSF model for the late winter/pre-calving season for boreal woodland caribou in the area of the lower Mackenzie River Valley, Northwest Territories.

95% confidence interval Variable ∃ coefficient Lower Upper Open conifer lichen 1.290 1.202 1.378 Open conifer other 0.192 0.105 0.280 Closed deciduous -2.242 -3.380 -1.103 Closed mixed needleleaf deciduous -1.326 -1.798 -0.855 Open mixed needleleaf deciduous 0.257 0.079 0.434 Tall shrub -0.372 -0.519 -0.225 Dwarf shrub -2.514 -4.482 -0.546 Ripparian 1.045 0.932 1.157 Water 0.330 0.195 0.465 Sparse/non-vegetated -0.734 -1.407 -0.061 Recent burn -0.887 -1.089 -0.684 Elevation -0.001 -0.001 -0.001 Slope -0.027 -0.046 -0.007 Aspect: flat -0.685 -0.832 -0.538 Aspect: north -0.480 -0.600 -0.361 Aspect: east -0.547 -0.665 -0.429 Aspect: west -0.295 -0.405 -0.185

35 Table 11. Proportions of the core and regional study area in the 5 relative probability of use categories and the percent of caribou locations found within these areas.

Relative Availability Percent of Probability Core study Regional study caribou use Season of Use (% of area) (% of area) locations

Calving/post-calving low 11 16 1 low-moderate 15 17 6 moderate 19 20 11 moderate-high 17 16 19 high 38 32 63

Early/mid-summer low 5 9 1 low-moderate 8 11 1 moderate 17 21 10 moderate-high 28 28 27 high 41 31 61

Late summer/fall low 18 22 3 low-moderate 7 9 3 moderate 9 10 5 moderate-high 20 18 18 high 47 40 70

Early winter low 3 5 1 low-moderate 4 5 2 moderate 8 9 4 moderate-high 37 36 25 high 48 45 69

Mid-winter low 6 11 0 low-moderate 21 18 10 moderate 22 20 13 moderate-high 14 15 11 high 38 36 65

Late winter/pre-calving low 3 5 0 low-moderate 12 11 3 moderate 15 16 7 moderate-high 27 27 22 high 42 41 68

36